-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmoreData.py
153 lines (136 loc) · 5.43 KB
/
moreData.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
import cv2
from os import listdir
from os.path import isfile, join
from os import rename
from matplotlib import pyplot as plt
from random import seed
from random import randint
from random import random
import numpy as np
from PIL import Image
import math
# Global variables #
#------------------------------------------------------------------------------#
baseDataset = "testDataset/GAN_DATASET/" # Root folder of the dataset
directories = {"david":["alpha_david/","alpha_david/","trimap_david/"],
"people":["justPeople/","justPeople_alpha/","justPeople_trimap/"]} #For each difirent dataset, indicate the alpha, trimap and foreground paths
outputFolder = "testDataset/DATASET/"
backgrounds = "bgs_europe/" # Folder containing the bg for composite
intermediate = "testDataset/intermediate/" # If you want to save intermediate step
random_bg = 0 # 1: random bg selection | 0: sequential bg selection
intSteps = 1 # Flag for intermediate step save
# Functions #
#------------------------------------------------------------------------------#
def addPadding(image, factor):
height, width = image.shape[:2]
if(height > width):
newWidth = round(factor*height)
fixed_image = cv2.copyMakeBorder( image, 0, 0, round((newWidth-width)/2), round((newWidth-width)/2), 0)
return fixed_image
else:
return image
#https://stackoverflow.com/questions/5789239/calculate-largest-rectangle-in-a-rotated-rectangle#7519376
def rotatedRectWithMaxArea(w, h, angle):
if w <= 0 or h <= 0:
return 0,0
width_is_longer = w >= h
side_long, side_short = (w,h) if width_is_longer else (h,w)
sin_a, cos_a = abs(math.sin(angle)), abs(math.cos(angle))
if side_short <= 2.*sin_a*cos_a*side_long or abs(sin_a-cos_a) < 1e-10:
x = 0.5*side_short
wr,hr = (x/sin_a,x/cos_a) if width_is_longer else (x/cos_a,x/sin_a)
else:
cos_2a = cos_a*cos_a - sin_a*sin_a
wr,hr = (w*cos_a - h*sin_a)/cos_2a, (h*cos_a - w*sin_a)/cos_2a
return round(wr/2),round(hr/2)
def randomRotation(fg_image,alpha_image,trimap_image):
height, width = fg_image.shape[:2]
center = (width/2, height/2)
r_angle=randint(-15,15)
rotate_matrix = cv2.getRotationMatrix2D(center=center, angle=r_angle, scale=1)
r_fg = cv2.warpAffine(src=fg_image, M=rotate_matrix, dsize=(width, height))
r_alpha = cv2.warpAffine(src=alpha_image, M=rotate_matrix, dsize=(width, height))
r_trimap = cv2.warpAffine(src=trimap_image, M=rotate_matrix, dsize=(width, height))
# Crop in Biggest rectangle
w_p,h_p = rotatedRectWithMaxArea(width,height,math.radians(r_angle))
c_h = round(height/2)
c_w = round(width/2)
cr_fg = r_fg[c_h-h_p:c_h+h_p, c_w - w_p:c_w+w_p]
cr_alpha = r_alpha[c_h-h_p:c_h+h_p, c_w - w_p:c_w+w_p]
cr_trimap = r_trimap[c_h-h_p:c_h+h_p, c_w - w_p:c_w+w_p]
#Intermediate save
if(intSteps):
cv2.imwrite(intermediate+"original.png",fg_image)
cv2.imwrite(intermediate+"rotation.png",r_fg)
cv2.imwrite(intermediate+"biggest_rectangle.png",cr_fg)
return [cr_fg,cr_alpha,cr_trimap]
def randomHUE(image):
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
h,s,v = cv2.split(hsv)
h_change = round((2*random() -1)*180)
# modify hue channel by adding difference and modulo 180
hnew = np.mod(h + h_change, 180).astype(np.uint8)
# recombine channels
hsv_new = cv2.merge([hnew,s,v])
# convert back to bgr
bgr_new = cv2.cvtColor(hsv_new, cv2.COLOR_HSV2BGR)
if(intSteps):
cv2.imwrite(intermediate+"hue.png",bgr_new)
return bgr_new
def randomComposite(bgs,fg,alpha,cont,random_bg):
if(random_bg):
bg_index = randint(0, len(bgs)-1)
else:
bg_index = cont%len(bgs)
bg = cv2.imread(baseDataset+backgrounds+bgs[bg_index])
h,w = fg.shape[:2]
crop_bg = cv2.resize(bg,(w,h))
# Composite
composite = np.zeros(fg.shape, dtype = fg.dtype)
for i in range(3):
composite[:,:,i] = alpha[:,:]/255.*fg[:,:,i] + (1 - alpha[:,:]/255.)*crop_bg[:,:,i]
return composite
def expandDataset():
cont = 0 #for names
#load BG
bg = [f for f in listdir(baseDataset+backgrounds) if isfile(join(baseDataset+backgrounds,f))]
#Read every directory
for dir in directories:
d_fg,d_alpha,d_trimap = directories[dir]
fg = [f for f in listdir(baseDataset+d_fg) if isfile(join(baseDataset+d_fg,f))]
for picture in fg:
for i in range(9):
# Load foreground, alpha and trimap
fg_image = cv2.imread(baseDataset+d_fg+picture)
if(dir == "david"):
alpha_image = cv2.imread(baseDataset+d_alpha+picture,cv2.IMREAD_UNCHANGED)[:,:,3]
else:
alpha_image = cv2.imread(baseDataset+d_alpha+picture,cv2.IMREAD_UNCHANGED)
trimap_image = cv2.imread(baseDataset+d_trimap+picture)
if(i == 0):
r_alpha = alpha_image
r_trimap = trimap_image
h_fg = fg_image
else:
r_fg,r_alpha,r_trimap = randomRotation(fg_image,alpha_image,trimap_image)
h_fg = randomHUE(r_fg)
h_fg = addPadding(h_fg,2)
r_alpha = addPadding(r_alpha,2)
r_trimap = addPadding(r_trimap,2)
c_image = randomComposite(bg,h_fg,r_alpha,cont,random_bg)
# Resize for entering the net
rc_image = cv2.resize(c_image,(256,128),cv2.INTER_AREA)
r_trimap = cv2.resize(r_trimap,(256,128),cv2.INTER_AREA)
r_alpha = cv2.resize(r_alpha,(256,128),cv2.INTER_AREA)
if(intSteps):
cv2.imwrite(intermediate+"composited.png",rc_image)
else:
#Save images
# --> composite
cv2.imwrite(outputFolder+"train_A/"+str(cont)+".png",rc_image)
# --> trimaps
cv2.imwrite(outputFolder+"train_B/"+str(cont)+".png",r_trimap)
# --> alphas
cv2.imwrite(outputFolder+"train_alphas/"+str(cont)+".png",r_alpha)
cont +=1
expandDataset()